Move resnet video models to single location (#1190)

* [WIP] Minor cleanups on R3d

* Move all models to video/resnet.py

* Remove old files

* Make tests less memory intensive

* Lint

* Fix typo and add pretraing arg to training script

Author: fmassa

references/video_classification/train.py

@@ -201,8 +201,7 @@ def main(args):
         pin_memory=True, collate_fn=collate_fn)
 
     print("Creating model")
-    # model = torchvision.models.video.__dict__[args.model](pretrained=args.pretrained)
-    model = torchvision.models.video.__dict__[args.model]()
+    model = torchvision.models.video.__dict__[args.model](pretrained=args.pretrained)
     model.to(device)
     if args.distributed and args.sync_bn:
         model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)

test/test_models.py

@@ -61,7 +61,7 @@ def _test_detection_model(self, name):
     def _test_video_model(self, name):
         # the default input shape is
         # bs * num_channels * clip_len * h *w
-        input_shape = (1, 3, 8, 112, 112)
+        input_shape = (1, 3, 4, 112, 112)
         # test both basicblock and Bottleneck
         model = models.video.__dict__[name](num_classes=50)
         x = torch.rand(input_shape)
@@ -145,6 +145,7 @@ def do_test(self, model_name=model_name):
 
     setattr(Tester, "test_" + model_name, do_test)
 
+
 for model_name in get_available_video_models():
 
     def do_test(self, model_name=model_name):

torchvision/models/video/__init__.py

@@ -1,3 +1 @@
-from .r3d import *
-from .r2plus1d import *
-from .mixed_conv import *
+from .resnet import *